Real-time classification of tyre models in high-performance vehicles: Comparing model-based and learning-based approaches

In the automotive scenario, automatic real-time tyre identification is paramount to optimize vehicle performance, enhance safety, and reduce consumption. Considering racing applications, tyre' classification supports strategic decisions, ensuring that the proper ones have been installed, while in urban scenarios, tyre recognition is crucial for in-vehicle control strategies to dynamically adjust their response, thus enhancing stability and safety during critical maneuvers. However, a clear gap has been identified in the literature on tyre classification. In fact, most existing approaches focus on usage monitoring such as tyre wear estimation while those explicitly addressing tyre type prediction are still lacking. Moreover, even considering usage monitoring approaches only, they face a trade-off between practicality and interpretability, a challenge that tyre classification methods are also expected to encounter. In fact model-based methods provide interpretable results but rely on scarcely available measurements such as tyre pressure and temperature. Conversely, machine learning approaches exploit easily accessible vehicle data and achieve high predictive performance, yet their limited interpretability constrains real-world applicability.To address this challenge, we present and compare two approaches for real-time tyre classification: a model-based method tailored to use more accessible vehicle measurements, and a learning-based approach designed with a focus on interpretability. To evaluate their performance, both a simulation scenario and real-world data are considered. While both models achieve optimal performance in simulation, data variability and noise in real-world conditions reduce the accuracy of the model-based approach. In contrast, the learning-based approach maintains an F1 score of 96.5% proving to be a practical and interpretable solution for real-time tyre recognition.